* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Garrett 03-30-2012 1 • Interlude: Calculus on spheres: invariant integrals, invariant
Polynomial ring wikipedia , lookup
Factorization wikipedia , lookup
Matrix calculus wikipedia , lookup
Cartesian tensor wikipedia , lookup
Factorization of polynomials over finite fields wikipedia , lookup
Hilbert space wikipedia , lookup
Fundamental theorem of algebra wikipedia , lookup
Group action wikipedia , lookup
Linear algebra wikipedia , lookup
Vector space wikipedia , lookup
Eisenstein's criterion wikipedia , lookup
Four-vector wikipedia , lookup
Covering space wikipedia , lookup
Bra–ket notation wikipedia , lookup
Basis (linear algebra) wikipedia , lookup
Homomorphism wikipedia , lookup
Garrett 03-30-2012 • Interlude: Calculus on spheres: invariant integrals, invariant ∆ = ∆S , integration-by-parts, etc. Decomposition of L2 (S n−1 ) into ∆S -eigenfunctions. Representation theory of orthogonal groups O(n, R) or SO(n, R). ... combine to prove Hecke’s identity: For a homogeneous, degree d harmonic 2 polynomial P on n , P (x) e−π|x| is a Fourier transform −d eigenfunction with eigenvalue i : R P (x) e−π|x| 2 b(ξ) = i−d · P (ξ) e−π|ξ| We need the representation theory of SO(n, R) 2 1 Garrett 03-30-2012 2 A little representation theory: Given a (topological) group o G, a group homomorphism T : G → AutC (V ) to the group of continuous linear automorphisms of a complex vector space V is a representation of G (on V ). The representation is finitedimensional when V is. When the group homomorphism G → AutoC (V ) is understood, the standard abuse is to say that V itself is the representation, not naming the homomorphism. In that case, rather than writing (T g)(v) for g ∈ G and v ∈ V , write simply gv for the action of g on v. C There is a continuity requirement, that G × V −→ V is continuous For finite-dimensional V , there is a (provably) unique (topological vector space) topology, so need not be specified explicitly. Topologies on infinite-dimensional V must be specified. The topology on G should be clear from context. Garrett 03-30-2012 3 A G-subrepresentation of a representation V of G is a complex vector subspace W of V stable under the action of G... and when V is infinite-dimensional W must be topologically closed. A representation V of G is irreducible if there is no proper Gsubrepresentation, that is, if there is no G-subrepresentation of V other than {0} and V itself. A G-homomorphism from one G-representation V to another Grepresentation W is a complex-linear map ϕ : V → W commuting with the action of G in the sense that ϕ(gv) = g ϕ(v) (for g ∈ G, v ∈ V ) Such G-homomorphisms are also called G-morphisms, G-maps, or also G-intertwinings, or G-intertwining operators. The collection of all G-intertwinings of V to W is HomG (V, W ). Garrett 03-30-2012 4 Theorem: (instance of Schur’s Lemma) For a finite-dimensional irreducible representation V of a group G, any G-intertwining ϕ : V → V of V to itself is scalar. Proof: First, claim that the collection HomG (V, V ) of all Gintertwinings of finite-dimensional V to itself is a division ring. Indeed, given ϕ 6= 0 in the ring HomG (V, V ), the image ϕ(V ) is readily seen to be a G-subrepresentation of V . For ϕ not the zero map, since V is irreducible, the image is either {0} or V , so must be V since ϕ 6= 0, and ϕ is surjective. Similarly, the kernel of ϕ is a G-subrepresentation, so is either the whole V , impossible since ϕ is not the zero map, or is {0}. Thus, ϕ is injective. Thus, ϕ is a bijection, and therefore has an inverse (easily seen to be a G-map). Thus, non-zero elements of the ring HomG (V, V ) have multiplicative inverses, and HomG (V, V ) is a division ring. Garrett 03-30-2012 5 For V finite-dimensional the whole collection of complex-linear endomorphisms of V is finite-dimensional. Certainly naturally lies inside the center of HomG (V, V ). For ϕ ∈ HomG (V, V ), the collection of rational expressions (ϕ) is a field, and is finitedimensional as a vector space over the copy of lying in the center of HomG (V, V ), so is algebraic. But is algebraically /// closed (by Liouville’s theorem), so ϕ ∈ . C C C C C Theorem: The space Hd of harmonic homogeneous total-degree d polynomials is an irreducible SO(n, )-representation. R R Proof: Suppose, to the contrary, that X is a proper SO(n, )subspace of Hd . Then the orthogonal complement Y of X inside Hd with respect to the SO(n, )-invariant inner product is also SO(n, )-stable. R R Garrett 03-30-2012 6 The subspace X consists of continuous functions, so, for x ∈ S n−1 , the functional f → f (x) is a linear functional on X. By an especially easy case of Riesz-Fischer, there is Fx ∈ X such that f (x) = hf, Fx i for all f ∈ X. Since X is rotation-invariant and not {0}, the functional f → f (x) cannot be 0 on all of X, so Fx 6= 0 ∈ X. Similarly, there is 0 6= Φx ∈ Y such that f (x) = hf, Φx i for all f ∈Y. By rotating, without loss of generality x = en . Then the following funny lemma proves that Fx and Φx must be scalar multiples of each other, contradiction. /// Garrett 03-30-2012 Let SO(n − 1) be the smaller orthogonal group which is the isotropy group of the point en = (0, . . . , 0, 1). R Lemma: On n , for each fixed d, in Hd there is a unique (up to constant multiples) SO(n − 1, )-fixed vector f , that is, so that h · f = f for every h ∈ SO(n − 1, ). R R Proof: A function invariant under SO(n − 1) must be of the form fo (x) = F (ρ2 , t) (where ρ2 = x21 + . . . + x2n−1 and t = xn ) Computing, ∆f0 = P i P ∂ 2 (2xi F1 ) + F22 = i (2F1 + 4xi F11 ) + F22 ∂xi = 2(n − 1)F1 + 4ρ2 F11 + F22 where subscripts denote derivatives. 7 Garrett 03-30-2012 8 Write the function as a polynomial in t with coefficients functions of ρ. These coefficients are necessarily homogeneous functions of ρ, so are powers of ρ. Write these as powers of ρ2 . fo (x) = F (ρ2 , t) = cd td + cd−1 td−1 (ρ2 )1/2 + cd−2 td−2 (ρ2 )1 + . . . + co (ρ2 )d/2 The harmonic-ness 0 = 2(n − 1)F1 + 4ρ2 F11 + F22 rewritten in powers of t gives a recurrence for the coefficients ci . Explicitly, looking at the (i − 2)th power of t in the harmonic-ness condition, ... Garrett 03-30-2012 9 ∂ 2 (d−i+2)/2 ci−2 (ρ ) 0 = 2(n − 1) ∂(ρ2 ) ∂ 2 2 2 (d−i+2)/2 +4ρ ci−2 (ρ ) i(i − 1)ci (ρ2 )(d−i)/2 2 ∂(ρ ) i h 1 1 1 = 2(n − 1) (d − i + 2) + 4 · (d − i + 2) (d − i) (ρ2 )(d−i)/2 ci−2 2 2 2 +i(i − 1)(ρ2 )(d−i)/2 ci Thus, i(i − 1)ci = −(d − i + 2)(n − 1 + d − i)ci−2 and the ci are determined completely from c1 and co . Garrett 03-30-2012 10 On the other hand, looking at the td−1 term in the harmonic-ness relation, 1 1 1 0 = [2(n − 1) (ρ2 )−1/2 + 4ρ2 (− )(ρ2 )−3/2 ]c1 = [(n − 1) − 1]ρ 2 2 2 So unless n = 2 we have c1 = 0, and all coefficients are determined by co , giving the desired uniqueness result. The case n = 2 can be treated more directly, from the easily demonstrable fact that Hd = C · (x + iy)d ⊕ C · (x − iy)d (on R2 ) This proves the uniqueness lemma, and the irreducibility of Hd as /// SO(n, ) representation. R Garrett 03-30-2012 11 The irreducibility of Hd is the key point, but there are a few other small requirements before Hecke’s identity is completed. As expected, two G-representations V, W are isomorphic when there is a vector space isomorphism V → W that is a G-hom. When the vector spaces are infinite-dimensional, the map is required to be a topological vector space isomorphism, as expected. Similar to Schur’s Lemma: Proposition: For non-isomorphic irreducible finite-dimensional G-representations V, W , the only G-hom ϕ : V → W is the zero map. Proof: The kernel and image of ϕ are G-subrepresentations, so are either the whole space or {0}. The case that ϕ : V → W is a vector space isomorphism is excluded by the non-isomorphism assumption. /// Garrett 03-30-2012 12 Proposition: (Complete Reducibility) A finite-dimensional G representation V with a G-invariant inner product h, i is a finite orthogonal direct sum of irreducible G-subrepresentations. Remark: A representation space with a G-invariant inner product is said to be unitary. Proof: Induction on dimension. For V irreducible, we are done. Otherwise, V has a proper subrepresentation W , which is a direct sum of irreducibles, by induction. The orthogonal complement W ⊥ of W is immediately G-stable, so is a G-subrepresentation, and also is a direct sum of irreducibles, by induction. /// Remark: There is no straightforward infinite-dimensional analogue of the previous, even for Hilbert spaces. The orthogonality argument still succeeds, but induction fails. Garrett 03-30-2012 The idea is that the SO(n, # : Hd −→ 13 R)-map C[x]≤d ≈ Hd ⊕ (other irreducibles) must map Hd to the other copy of Hd , and, by Schur’s lemma, be a scalar. This idea is correct, but has not quite been proven so far. Specifically, although we easily showed that, for irreducibles σ, τ , the space of G-homs HomG (σ, τ ) is either 0 or depending on whether σ 6≈ τ or σ ≈ τ , this does not instantly address the question of G-maps to sums of irreducibles. C It is best to examines some clarifying structure. Garrett 03-30-2012 14 Isotypes and co-isotypes A direct sum of a number of copies of an irreducible π is denoted m·π = π . . ⊕ π} | ⊕ .{z m Given an irreducible π of G, we want to specify a G-sub V π of a G-rep’n V such that any map m · π → V factors (uniquely) through V π , that is, m · π → V π ⊂ V , the π-isotype of V . Dually, the π-co-isotype Vπ of V is the quotient of V such that any map V → m · π factors through Vπ : V → Vπ → m · π. A priori, existence is unclear, but on categorical grounds they are unique up to unique isomorphism if they do exist at all. For unitary representations, the kernel of the map to the coisotype has an orthogonal complement, so the co-isotype is naturally isomorphic to a sub-object, ... but in general we should not expect this simplicity. Garrett 03-30-2012 15 Happily, for finite-dimensional irreducibles π of compact G, there is a natural projector to the π-isotype. It is not obvious, but the history of these issues does reasonably lead to the following. For a finite-dimensional irreducible π of compact G, the character χπ of G is a function on G defined by χπ (g) = trace π(g) Proposition: For any G-representation V , the map Z v −→ χπ · v = χπ (g) g v dg G is a G-hom projecting V → V π , where π ∨ is the contragredient (dual) representation of π.